摘要
设计了一套基于人脸表情特征的情感交互系统(情感虚拟人),关键技术分别为情感识别、情感计算、情感合成与输出三个方面。情感识别部分首先采用特征块的方法对面部静态表情图形进行预处理,然后利用二维主元分析(2DPCA)提取特征,最后利用多级量子神经网络分类器实现七类表情识别分类;在情感计算部分建立了隐马尔可夫情感模型(HMM),并且用改进的遗传算法估计模型中的参数;在情感合成与输出阶段,首先采用NURBS曲面和面片相结合的算法,建立人脸三维网格模型,然后采用关键帧技术,实现了符合人类行为规律的连续表情动画。最后完成了基于人脸表情特征的情感交互系统的设计。
This paper presented a systematic design method for the emotion interactive system （ virtual emotional human） based on the characteristics of facial expression. The key technologies involved emotion recognition, affective computing, emotion synthesis and output. In the part of emotion recognition, firstly, processed the static facial expression by the method of characteristic blocks, then the method 2DPCA was used to extract characteristics, finally applicated the muhi-level quantum neural network classifiers in achieving seven-class classification of expression. In the part of affective computing, built a hidden Markov model, and estimated the model parameters with improved genetic algorithm. In the part of emotion synthesis and output, establish a three-dimensional mesh face, and then used key frame techniques to achieve a continuous pattern of behavior consistent with human facial animation.
出处
《计算机应用研究》
CSCD
北大核心
2012年第3期1111-1115,共5页
Application Research of Computers
基金
国家自然科学基金资助项目(60973095)
关键词
情感虚拟人
二维主元分析
多级量子神经网络
隐马尔可夫情感模型
virtual emotional human
two dimension principal component analysis （2DPCA）
muhi-layer quantum neural network
hidden Markov model （HMM）
